Friday, August 7th 2009

Gulftown Seamlessly Compatible with X58 Chipset Platform

Nearly an year away from launch, Intel's 6-core Gulftown processor is creating buzz in the media. The new processor which has already been sampled, and tested, some of its first documentation has been leaked to the media in the form of a slide that lists out some details. To begin with, Gulftown will be the first high-performance processor built on the 32 nm second-generation HKMG process. Intel's first commercial 32 nm based processors are dual-core chips which will sell alongside the much larger 45 nm CPU lineup. Next of-course, is that it has six processing cores with HyperThreading enabling 12 logical CPUs. With TurboBoost, the processor powers down inactive cores and overclocks the active ones, thus boosting performance on single/few-threaded apps while reducing power consumption.

Next up is its integrated memory controller supporting DDR3-1066 memory. The same standard is listed for current Core i7 processors, though they are proven to work at higher memory bus speeds. The part that perhaps matters the most is that the processor is seamlessly compatible with the X58 chipset, is based on socket LGA-1366, and has the same TDP rating as Core i7 900 series processors: 130W. Existing motherboards that handle Core i7 processors are technically capable to handle Gulftown. The processor will only be available in the high-end (enthusiast) category, and according to the latest roadmaps, is slated for Q2 2010.
Source: DonanimHaber
Add your own comment

38 Comments on Gulftown Seamlessly Compatible with X58 Chipset Platform

#26
Wile E
Power User
FordGT90ConceptIt is. Nehalem sucks at games--so did Pentium 4. Nehalem has hyperthreading--so did Pentium 4. Nehalem kicks ass at multimedia--so did Pentium 4. Nehalem runs hot--so did Pentium 4. Nehalem codename was supposed to be a Pentium 4 (~2000) with Common System Interface (CSI). CSI is now known as "QuickPath Interface." Nehalem was scraped and revived many times until about a year prior to the release of Core 2. Everything about Nehalem screams Netburst.


There's only two things different about Nehalem that wasn't there on the original Netburst codename: multi-core and turbo mode. Multi-core because the dual- and quad-core revolution didn't start until 2005 and turbo mode was a revival from the 286, 386, and 486. I have no idea why they did. Perhaps it is because Nehalem has a core clock of 133 MHz so 1x or 2x more on the multiplier isn't going to hurt anything.
No, architecturally, it has nothing to do with P4. The only thing they have in common is Hyper Threading, but Nehalem's version is actually completely different in operation, and some code names. Nehalem is still based heavily on Core 2, which gets its roots from P3. Reviving the idea of an interconnect like "CSI", or the use of multiple threads from a single core, does not make it related to netburst. It just means the engineering department recycled some terms.

We wouldn't be seeing the IPC we are if it's roots were in netburst. Even with HTT disabled, it's IPC is higher then all other current x86 chips out. Netburst failed miserably at ipc, due to it's retarded long and narrow pipelines that favored clock speed over ipc whick took forever to flush in case of a cache miss or prediction error (which was what made it the netburst architecture.) Nehalem does not have long and narrow pipes. Even with todays much improved branch prediction, netburst still couldn't be made this efficient.

And it does not suck at gaming. Where in the hell do you get that conclusion? Just because it doesn't benefit gaming, doesn't mean it sucks at it. Almost all games on the market are gpu limited. The cpu makes no difference to gameplay (within reason, of course).
Posted on Reply
#27
FordGT90Concept
"I go fast!1!11!1!"
entropy13So basically when I upgraded from a Pentium 4 560 (Prescott) to a Core i7 920 I just "modernized" more or less? :laugh:
Pretty much--from a broken Netburst to a properly functioning Netburst. ;)



Netburst had serious leakage and cache miss issues. Nehalem has the leakage and caching issues fixed.

Penryn has 14 stages. Early Netburst processors have 20 stages. Later Netburst processors (Prescott) have 32 stages. Guesstimates put Nehalem at somewhere between 20 and 24 stages.

The SMT capabilties of Netburst were strapped on after they realized how much of the processor sat idle. Nehalem was designed knowing SMT was going to be included so it is a much better implementation.

Nehalem is effectively Netburst done right. Remember, Intel spent over 7 years trying to fix Netburst and never really succeeded.
Wile EAnd it does not suck at gaming. Where in the hell do you get that conclusion? Just because it doesn't benefit gaming, doesn't mean it sucks at it. Almost all games on the market are gpu limited. The cpu makes no difference to gameplay (within reason, of course).
At low resolutions where the CPU is the most important, Nehalem does better. The higher the resolution, the worse Nehalem does compared to Core 2 and Phenom II. I've never found an explaination for this--perhaps a shortcoming of QPI.
Posted on Reply
#28
Wile E
Power User
FordGT90ConceptPretty much--from a broken Netburst to a properly functioning Netburst. ;)



Netburst had serious leakage and cache miss issues. Nehalem has the leakage and caching issues fixed.

Penryn has 14 stages. Early Netburst processors have 20 stages. Later Netburst processors (Prescott) have 32 stages. Guesstimates put Nehalem at somewhere between 20 and 24 stages.

The SMT capabilties of Netburst were strapped on after they realized how much of the processor sat idle. Nehalem was designed knowing SMT was going to be included so it is a much better implementation.

Nehalem is effectively Netburst done right. Remember, Intel spent over 7 years trying to fix Netburst and never really succeeded.



At low resolutions where the CPU is the most important, Nehalem does better. The higher the resolution, the worse Nehalem does compared to Core 2 and Phenom II. I've never found an explaination for this--perhaps a shortcoming of QPI.
Again, having similarities to netburst, does not make it a derivative of netburst.

And all the tests I've seen show i7 slightly ahead at higher resolutions in most games, with only a couple of exceptions. Even tho i7 pulls ahead slightly, it makes no real difference anyway, as the playability remains the same between all the chips at high res gaming. Thus, no modern chips actually "suck" at gaming.
Posted on Reply
#29
FordGT90Concept
"I go fast!1!11!1!"
Wile EAgain, having similarities to netburst, does not make it a derivative of netburst.
Does having a child look like you not make it your child? Unfortunately, processors don't have DNA to test.
Posted on Reply
#30
Wile E
Power User
FordGT90ConceptDoes having a child look like you not make it your child? Unfortunately, processors don't have DNA to test.
My nephew looks like me, but is not my child.
Posted on Reply
#31
FordGT90Concept
"I go fast!1!11!1!"
You and him are both humans making you 99.900% alike, genetically. Considering heredity, you are no less than 99.925% alike.

Shall we stick a percentage on how alike Pentium 3, Pentium 4, Core 2, and Core i7 are? It might be fun. :D
Posted on Reply
#32
Wile E
Power User
FordGT90ConceptYou and him are both humans making you 99.900% alike, genetically. Considering heredity, you are no less than 99.925% alike.

Shall we stick a percentage on how alike Pentium 3, Pentium 4, Core 2, and Core i7 are? It might be fun. :D
Go right ahead.
Posted on Reply
#33
FordGT90Concept
"I go fast!1!11!1!"
But...what definies the architecture? SMT for sure does. I'd argue that SSE doesn't because such technologies can be integrated into pretty much any x86 architecture. x86-64 does...

Maybe MCM... but multi-core really isn't an architectural feature.
Posted on Reply
#34
Wile E
Power User
FordGT90ConceptBut...what definies the architecture? SMT for sure does. I'd argue that SSE doesn't because such technologies can be integrated into pretty much any x86 architecture. x86-64 does...

Maybe MCM... but multi-core really isn't an architectural feature.
SMT doesn't define it, if it's designed in a completely different manner than the original SMT. Only the concept itself remains, not the design.
Posted on Reply
#35
FordGT90Concept
"I go fast!1!11!1!"
So, then, what does? XD

I'm down to just x86-64/x86 now. :(
Posted on Reply
#36
Wile E
Power User
FordGT90ConceptSo, then, what does? XD

I'm down to just x86-64/x86 now. :(
That's what I mean. There's nothing to link it directly to netburst, aside from a longer pipe than Core2. Likely, it's a derivative of both.
Posted on Reply
#37
FordGT90Concept
"I go fast!1!11!1!"
It is my belief that it borrows more from Netburst than it does from P6. At the same time, I think they may have started with a clean slate and aimed for what the original Nehalem had promised (high clock speeds, SMT, and CSI).
Posted on Reply
Add your own comment
May 4th, 2024 16:08 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts